Endmember selection techniques for improved spectral unmixing

Abstract

Hyperspectral imaging provides the potential to derive sub-pixel material abundances. This has significant utility in the detection of sub-pixel targets or targets concealed under canopy. The linear mixture model describes spectral data in terms of a basis set of pure material spectra or endmembers. The success of such a model is dependent on the choice and number of endmembers used and the unmixing process. Endmember spectra may come from field or laboratory measurements, however, differences between sensors and changes in environmental conditions may mean that the measurement is not representative of the material as found in the scene. Alternatively, a number of algorithms exist to select spectra from the data directly, but these assume pure examples of the complete set of materials exist within the imagery. In either case, the chosen set of endmembers may not optimally describe the data in a linear mixing sense. In this paper some new methods for endmember selection are presented. These are evaluated on hyperspectral imagery and the results compared with those of a well-known automatic selection technique. Finally, an improved unmixing architecture is proposed which is self-consistent in terms of endmember selection and the unmixing process.

Deforestation in the Miombo woodlands: a pixel-based semi-automated change detection method

“…It is generally recommended to validate the N-FINDR results against approaches like the PPI. After positive validation results, the N-FINDR algorithm has significant advantages over manual endmember retrieval, or the usage of reference spectra of surface material as discussed in Howes et al (2004).…”

“…As a constraint, all three abundance values (F i ) have to be >0 and <1. This is known as the non-negativity constraint in previously published literature (Howes et al 2004). Additionally, the sum of the fractions has to be one (sum to one constraint; Howes et al 2004).…”

“…The final simplex and endmembers of the scene from 1986 describe 99.49% of all enclosed pixels, while those from the scene of 1998 describe 99.35%. An excellent summary of recent endmember determination methods is provided by Howes et al (2004).…”

The Successive Projection Algorithm (SPA), an Algorithm with a Spatial Constraint for the Automatic Search of Endmembers in Hyperspectral Data

“…bad pixels) as endmembers. Indeed Howes et al (2004) reported that convex-based endmember extraction methods were susceptible to outliers since only a single spurious pixel can significantly alter the endmember simplex [21]. Outliers may result for example from noise or atmospheric effects in the data.…”

“…Whether this assumption is met is in part a function of the nature of the scene (spatial arrangement of targets) and the spatial resolution of the imagery. In the case of SMACC, N-FINDER, VCA, SGA and Max-D, one pixel (the most extreme pixel) is selected to represent one endmember and thus these methods suffer from the inherent sensitivity of convex geometry to outlier pixels [21]. Although principal component analysis (PCA), minimum noise fraction transform (MNF) and singular value decomposition (SVD) can be applied to the data prior to the endmember extraction to minimize the impact of noise [14, 28], these can also result in the loss of detection of useful endmember pixels characterized by subtle spectral detail.…”